1 / 25

A Framework for Using Processor Cache as RAM (CAR) Eswaramoorthi Nallusamy

A Framework for Using Processor Cache as RAM (CAR) Eswaramoorthi Nallusamy University of New Mexico October 10, 2005. Acknowledgement. Prof. David A. Bader, University of New Mexico Dr. Ronald G. Minnich and his team, Advanced Computing Labs, Los Alamos National Laboratory. Overview.

lonato
Télécharger la présentation

A Framework for Using Processor Cache as RAM (CAR) Eswaramoorthi Nallusamy

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. A Framework for Using Processor Cache as RAM (CAR) Eswaramoorthi Nallusamy University of New Mexico October 10, 2005

  2. Acknowledgement Prof. David A. Bader, University of New Mexico Dr. Ronald G. Minnich and his team, Advanced Computing Labs, Los Alamos National Laboratory

  3. Overview • Processor’s cache-as-RAM (CAR) • Architectural Components affecting CAR • Cache-as-RAM initialization • Conclusion

  4. Processor’s Cache as RAM (CAR) • Using processor’s cache as RAM until the RAM is initialized • CAR allows the memory init code to be written in ANSI C and compiled using GCC • Solves all the limitations of previous solutions comprehensively • Porting LinuxBIOS to different motherboards is quick • Feasibility of Cache-as-RAM • Cache memory is ubiquitous across processor architectures & families • Cache subsystem can be initialized with few instructions • Cache init code doesn’t depend on any mother board components • Generally doesn’t require changes across a processor family • Can be ported easily among processor families

  5. Architectural Components affecting CAR • Protected Mode • Cache Subsystem • Inter-Processor Interrupt Mechanism • HyperThreading Feature

  6. Protected Mode • Native mode of operation • Provides rich set of architectural features including cache subsystem • Requires, at least, a Global Descriptor Table (GDT) with entries for Code Segment (CS), Data Segment (DS) & Stack Segment (SS) • Each GDT entry describes properties of a memory segment • Can be entered by setting Control Register 0 (CR0) flags

  7. Cache Subsystem • Different levels of caches (L1, L2) • Cache coherency maintained using Modified-Exclusive-Shared-Invalid (MESI) protocol • Cache control mechanisms • Cache Disable (CD) flag in CR0 – system wide • Memory Type Range Registers (MTRRs) – memory region • Page Attribute Table (PAT) Register – individual pages

  8. Cache Operating Modes

  9. Caching Methods • Uncacheable/Strong Uncacheable • Memory locations not cached • Write Combining • Memory locations not cached • Writes are combined to reduce memory traffic • Write Through • Memory locations are cached • Reads come from cache line; may cause cache fills • Writes update the system memory

  10. Caching Methods • Write Back • Memory locations are cached • Reads come from cache line; may cause cache fills • Writes update cache line • Modified cache lines written back only when cache lines need to de-allocated or when cache coherency need to be maintained • Write Protected • Memory locations are cached • Reads come from cache line; may cause fills • Writes update system memory and the cache line is invalidated on all processors

  11. Inter-Processor Interrupt (IPI) • Used for interrupt based communication among processors in a SMP system • Each processor is assigned a unique ID • IPI is raised by writing appropriate values into Interrupt Command Register (ICR)

  12. Interrupt Command Register (ICR)

  13. HyperThreading Feature

  14. CAR Principle of Operation • For the cache to act as RAM • The cache subsystem must retain the data in the CAR region irrespective of internal cache accesses falling within and outside the CAR region. (Assuming there is no external memory access activity) • Any memory accesses within the CAR region must not appear on the Host Bus as memory transactions.

  15. CAR Principle of Operation Contd… • Normal Cache Mode (CR0.CD=0) • Read/Write misses cause cache line fills • Cache lines replaced when there are no free cache line fills • No-fill Mode (CR0.CD=1) • Read/Write misses access memory • Cache lines are never replaced • An MTRR describing CAR region’s caching method is required • Only Write-Back caching method is capable of confining Read/Write access within cache subsystem

  16. CAR Principle of Operation Contd… • Thus with no system memory/internal cache activity in other processors the cache subsystem operating in No-Fill Mode having at least one Write-Back CAR region behaves as RAM. • As only the Normal Cache Mode is capable of filling the cache lines, we must fill the CAR region’s cache lines in Normal Cache Mode before entering No-Fill Mode. • Only the Boot Strap Processor’s (BSP) internal cache is initialized to behave as CAR.

  17. GDT and CR0 setup • An NVRAM based flat-model GDT to access the processor address space is setup during LinuxBIOS build process • GDTR is initialized to point to the GDT in NVRAM • BSP enters protected mode by setting CR0.PE flag

  18. MTRR setup

  19. MTRR setup Contd…

  20. Cache Subsystem Initialization • Establish valid tags for the cache lines in the CAR region • Enter Normal Cache Mode (CR0.CD=0) • Cache lines in CAR region are marked valid by reading memory locations within CAR region • Enter No-Fill Mode (CR0.CD=1) to avoid cache line fills/flushes • The cache lines are analogous to uninitialized memory. • Initialize the cache lines with some recognizable pattern for ease of debugging

  21. Cache Subsystem Contd…

  22. HyperThreaded Processor Initialization • For a HT enabled processor • Cache subsystem is common to both the logical processors • CR0 is unique to each logical processors • The ultimate cache operating mode is decided by the “OR” result of CR0.CD flag of each logical processors • Only one logical processor is designated as BSP using MP init protocol • A logical processor can’t directly access other logical processor’s resources • An IPI is used by the BSP to clear the CR0.CD flag of the logical processor • An IPI handler is provided to clear the logical processor’s CR0.CD

  23. HT Processor Initialization Contd…

  24. HT Processor Initialization Contd…

  25. Conclusion • Limitations of LinuxBIOS imposed by ROMCC is addressed • Memory Init code can be written in ANSI C and compiled using standard C compilers • Memory Init object code size is reduced by factor of Four

More Related